You usually use WNOHANG and WUNTRACED in different cases.
Case 1: Suppose you have a process which spawns off a bunch of children and needs to do other stuff while the children are running. These children sometimes exit or are killed, but the kernel will hold onto their exit status until some other process claims it via wait() or waitpid(). So, your parent process needs to call wait()/waitpid() on occasion to let the kernel rid itself of the remains of the child. But we don't want wait()/waitpid() to block, because, in this case, our process has other things that it needs to do. We just want to collect the status of a dead process if there are any. That's what WNOHANG is for. It prevents wait()/waitpid() from blocking so that your process can go on with other tasks. If a child died, its pid will be returned by wait()/waitpid() and your process can act on that. If nothing died, then the returned pid is 0.
Case 2: Suppose your parent process, instead, wants to do nothing while children are running. You don't want to just have it do some thumb-twidling for-loop, so you use a normal wait()/waitpid() without WNOHANG. Your process is taken out of the execution queue until one of the children dies. But what if one of your children is stopped via a SIGSTOP? Your child is no longer working on the task you have set it to, but the parent is still waiting. So, you've got a deadlock, in a sense, unless the child is continued by some means external to your parent and that child. WUNTRACED allows your parent to be returned from wait()/waitpid() if a child gets stopped as well as exiting or being killed. This way, your parent has a chance to send it a SIGCONT to continue it, kill it, assign its tasks to another child, whatever.
wait(2)
. – Paleozoic-1
, it doesn't wait for all children to finish, it waits for any child to finish. Once any child has finished, it is being reaped. – Medlock